China's AI Chip Makers Post Triple-Digit Revenue Growth

Chinese AI Chip Makers Deliver Explosive Q1 Results

China's domestic AI chip industry just posted its strongest quarter on record. Three leading chipmakers — Cambricon, Moore Threads, and Muxi — reported Q1 2026 revenue growth of 160%, 155%, and 75% respectively, signaling that the country's push for semiconductor self-sufficiency is gaining serious commercial traction.

The results arrive against a backdrop of escalating U.S. export controls on advanced AI chips, which have effectively locked Chinese companies out of accessing Nvidia's most powerful processors. Rather than slowing China's AI ambitions, the restrictions appear to be fueling a domestic chip boom that is reshaping the competitive landscape — not just within China, but globally.

Key Takeaways

• Cambricon led the pack with 160% year-over-year revenue growth in Q1 2026
• Moore Threads followed closely at 155% growth, confirming strong demand for GPU alternatives
• Muxi posted 75% growth, rounding out a uniformly positive quarter
• Huawei's Ascend 950PR processor launched in March, claiming 2.87x the inference performance of Nvidia's H20
• The competitive landscape is shifting from a 'free-for-all' to a clear market stratification among domestic players
• China's AI chip sector is entering what analysts call a phase of 'definitive growth' with unprecedented demand certainty

Huawei's Ascend 950PR Changes the Game

The single most important development in Q1 wasn't any individual earnings report — it was Huawei's formal launch of the Ascend 950PR processor in March. This chip represents Huawei's first processor deeply optimized specifically for AI inference workloads, a segment that is rapidly outpacing training in terms of commercial demand.

The specifications are noteworthy. The Ascend 950PR delivers 1.56 PFlops of FP4 precision compute on a single card, features 112GB of Huawei's proprietary HBM memory with 1.4 TB/s bandwidth, and operates at a 600W power envelope. Huawei debuted the chip on its new Atlas 350 AI training and inference accelerator card.

According to Zhang Dixuan, president of Huawei's Ascend computing business, the Atlas 350 delivers 2.87x the single-card performance of Nvidia's H20 — currently the most powerful AI chip Nvidia is permitted to sell in China. The Atlas 350 is also positioned as China's only inference product supporting FP4 low-precision computing, a capability that is becoming critical for deploying large language models cost-effectively at scale.

Huawei has also published a 3-year roadmap for its Ascend AI chip lineup, signaling sustained investment and a clear intent to establish the platform as China's de facto standard for AI computing infrastructure.

From Price War to Market Stratification

When the Q1 financial reports from Cambricon, Moore Threads, and Muxi are examined side by side, a structural shift becomes apparent. China's domestic AI chip market is transitioning from what industry observers describe as a chaotic 'melee' into a more defined competitive hierarchy.

This stratification matters for several reasons:

• Huawei is consolidating its position at the top of the stack, leveraging its end-to-end ecosystem from chips to cloud services
• Cambricon is carving out a strong position in the enterprise and government sectors, where its chips power inference workloads for major Chinese tech platforms
• Moore Threads is aggressively targeting the GPU-compatible computing market, appealing to developers who need CUDA-like programmability
• Muxi is building a niche in specific vertical applications, though its relatively slower growth rate suggests it may face pressure from larger rivals

The divergence in growth rates — ranging from 75% to 160% — is itself a signal. In the earliest stages of any market, all players tend to grow at similar rates as demand outstrips supply. When growth rates begin to diverge, it typically indicates that customers are making more deliberate choices about which platforms to standardize on.

Why This Matters Beyond China

For Western companies and investors, the rapid maturation of China's AI chip ecosystem carries significant implications. The prevailing assumption in Silicon Valley has been that U.S. export controls would maintain a multi-year technology gap between Chinese and Western AI capabilities. The Q1 results challenge that assumption.

The numbers suggest that Chinese AI chip demand is not merely surviving export restrictions — it is thriving because of them. Government procurement mandates, corporate patriotic purchasing, and genuine technological progress are all contributing to a demand environment that gives domestic chipmakers unusual revenue visibility.

This has direct consequences for Nvidia, which has seen its China revenue decline substantially since export controls tightened. Every dollar flowing to Cambricon or Huawei's Ascend platform is a dollar that is unlikely to return to Nvidia even if export controls were hypothetically relaxed. Platform lock-in is real, and China's AI ecosystem is building its own.

For AMD and Intel, the window to compete in China's AI accelerator market is narrowing rapidly. Both companies have struggled to gain meaningful traction even in the Western market against Nvidia, and the domestic preference dynamics in China make the challenge exponentially harder.

The Inference Economy Drives Demand

A critical driver behind these Q1 results is the explosive growth of AI inference demand. While training large models requires enormous compute clusters, inference — running those trained models to serve actual users — requires even more total compute as AI applications scale to hundreds of millions of users.

China's AI application ecosystem has expanded dramatically in recent quarters. Companies like DeepSeek, Baidu, Alibaba, and ByteDance are deploying large language models across consumer and enterprise applications at massive scale. Each of these deployments requires inference hardware, and with Nvidia's most capable chips unavailable, domestic alternatives are absorbing the entirety of this demand.

The Ascend 950PR's focus on inference optimization reflects this market reality. Huawei is betting that the inference market will ultimately dwarf the training market in total chip revenue — a bet that aligns with projections from Western analysts at firms like Morgan Stanley and Goldman Sachs, who estimate that inference will account for 60-70% of total AI compute spending by 2027.

Technical Gaps Remain But Are Narrowing

It is important to maintain perspective on where Chinese AI chips stand relative to the global frontier. A direct comparison between the Ascend 950PR and Nvidia's latest Blackwell B200 architecture reveals that a meaningful performance gap persists, particularly in:

• Training efficiency for models above 100 billion parameters
• Software ecosystem maturity, where Nvidia's CUDA platform maintains a substantial lead
• Advanced packaging and manufacturing process technology
• Memory bandwidth at the system level for multi-node training clusters
• Developer tooling and third-party library support

However, the relevant comparison for Chinese customers is not against the B200 — which they cannot purchase — but against the H20, which is the most advanced chip available to them from Nvidia. On that benchmark, Huawei's claims of a 2.87x performance advantage are commercially decisive.

The software gap is also closing faster than many Western observers expected. Huawei's MindSpore framework and Cambricon's Neuware SDK have both made significant strides in compatibility and performance optimization. While neither matches CUDA's breadth, they are increasingly sufficient for the most common AI workloads.

Looking Ahead: What to Watch in Q2 and Beyond

The trajectory established in Q1 sets up several critical questions for the remainder of 2026. First, can these growth rates sustain? The 160% figure from Cambricon is extraordinary, but it comes off a relatively small revenue base. As these companies scale, maintaining triple-digit growth will become mathematically harder.

Second, Huawei's roadmap execution will be pivotal. The company has signaled aggressive timelines for next-generation Ascend chips, and any delays could create openings for rivals — both domestic and international.

Third, the U.S. policy environment remains a wildcard. Further tightening of export controls could accelerate China's domestic chip adoption even further, while any relaxation could introduce competitive pressure that tests whether domestic chips can win on merit alone.

For global technology leaders, the message from Q1 is clear: China's AI chip industry is no longer an aspiration — it is a commercial reality growing at triple-digit rates, backed by massive demand, and increasingly capable of meeting the needs of the world's second-largest AI market.